leaks, disclosures and internal communication · leaks, disclosures and internal communication...
TRANSCRIPT
Electronic copy available at: https://ssrn.com/abstract=2999835
Leaks, disclosures and internal communication∗
Snehal Banerjee and Taejin Kim
November 2017
Abstract
We study how leaks affect a firm’s communication decisions and real efficiency. A
privately informed manager strategically chooses both public disclosure and internal
communication to employees. Public disclosure is noisy, but in the absence of leaks,
internal communication is perfectly informative because it maximizes employee coor-
dination and efficiency. The possibility of public leaks distorts these choices: we show
that more leakage worsens internal communication and can reduce real efficiency, de-
spite increasing public disclosure. We discuss the implications of our results for recent
regulations that protect and encourage whistleblowers in financial markets.
Keywords: leaks, whistleblower, disclosure, internal communication, efficiency
∗Banerjee ([email protected]) is at the University of California, San Diego, and Kim ([email protected]) is at The Chinese University of Hong Kong. We thank Bradyn Breon-Drish, JesseDavis, Naveen Gondhi, Doron Levit and Sivan Frenkel for helpful comments and discussions. All errors areour own.
1
Electronic copy available at: https://ssrn.com/abstract=2999835
1 Introduction
Publicity is justly commended as a remedy for social and industrial diseases.
Sunlight is said to be the best of disinfectants; electric light the most efficient
policeman.
— Louis Brandeis, Other People’s Money – and How Bankers Use It (1914)
By encouraging public leaks of internal information, regulatory policy strives to deter wrong-
doing by firms. For example, the SEC’s Dodd-Frank Whistleblower Program is designed to
improve protection and provide incentives to whistleblowers in financial markets.1 Since in-
ception, the program has awarded more than $111 million to 34 whistleblowers, and has led
to SEC enforcement actions in which over $584 million in financial sanctions were ordered.
In fiscal year 2016 alone, the SEC’s Office of the Whistleblower received over 4,200 tips and
awarded over $57 million as part of the program.
While these policies are arguably effective at detecting unlawful behavior by market par-
ticipants, they also have unintended consequences. By increasing the likelihood that internal
communication is leaked publicly, these policies can distort how a firm’s employees commu-
nicate with each other within the firm and to market participants. Specifically, we show
that increasing the likelihood of leaks leads to more public disclosure, but coarser internal
communication. Moreover, the overall effect of more leakage on real (allocative) efficiency is
not monotonic, and depends on the ex-ante informativeness of internal communication.
We consider a setting where a manager, who is privately informed about the firm’s funda-
mentals, strategically chooses both public disclosure and internal communication. Because
public disclosures affect the short-term stock price, the manager has an incentive to with-
hold negative news about fundamentals from the public. However, she prefers her internal
communication to be as informative as possible, since this improves the extent to which
her employees can coordinate on the appropriate action for the firm. We assume that the
manager’s public disclosures are verifiable, but that internal communication is cheap talk.
We show that the possibility of leakage introduces a tradeoff between internal and external
communication. On the one hand, the benefit from withholding negative news from the
public is lower since it may be leaked anyways. On the other hand, the possibility of leaks
creates an incentive for the manager to distort her message to employees, which makes it
more difficult to sustain informative communication internally. As a result, more leakage
1The SEC Whistleblower Program is a recent addition to a number of other, similar financial mar-ket regulations (e.g., US Whistleblower Protection Act (1989), Section 806 of Sarbanes-Oxley Act (2002),Whistleblower Protection Enhancement Act (2007)). See the U.S. Securities and Exchange Commission, 2016Annual Report to Congress on The Dodd-Frank Whistleblower Program (https://www.sec.gov/files/owb-annual-report-2016.pdf) for more details.
2
leads to more informative public disclosure, but less informative internal communication.
In our model, real efficiency is driven by the degree of coordination among employees. As
a benchmark, we first consider the case when the manager’s incentives are aligned with those
of the employees. In this case, we show efficiency is maximized when there are no leaks: inter-
nal communication is perfectly informative, and as a result, coordination is maximized. An
increase in leakage has two offsetting effects on efficiency. On the one hand, internal commu-
nication is coarser, which reduces the information available to employees and, consequently,
makes coordination more difficult. On the other hand, there is more public disclosure, which
provides more information to employees and, thereby, improves coordination. Initially, the
first effect dominates and so efficiency decreases with leakage. However, when the probability
of leaks is sufficiently high, internal communication is uninformative and the second effect
takes over. As a result, when the likelihood of leakage is very high, real efficiency increases
with leakage.
We then extend the analysis to the case in which there are conflicts of interest between
the manager and the other stakeholders of the firm. We show that in this case, leakage
can have a disciplinary effect: efficiency can be higher in the presence of leaks than in the
absence of leaks. Because the manager’s incentives are not aligned with the employees’,
internal communication is coarse even in the absence of leakage. When the probability of
leakage is sufficiently high, the impact of the increased public disclosure offsets the decrease
due to worse internal communication, and overall efficiency can be higher as a result.
Our analysis suggests a novel channel through which regulatory policy towards whistle-
blowers can affect firm behavior, and offers a number of implications. A higher likelihood
of leakage should be associated with more public disclosure, but less informative internal
communication. These effects should be especially pronounced in firms where compensation
contracts over-weight short-term performance (e.g., through higher stock / option grants).
Policies that increase leakage are more likely to be beneficial for firms with severe conflicts
of interest and poor internal communication, but counter-productive for firms with good
governance and internal communication. Accounting for this cross-sectional variation when
evaluating the impact of regulatory policies is important in accurately measuring their ef-
fectiveness.
Our model builds on the seminal work by Crawford and Sobel (1982) on cheap talk and by
Dye (1985) and Jung and Kwon (1988) on strategic verifiable disclosure. Our paper is part of
a larger literature that studies communication in settings with a single sender and multiple
receivers (e.g., Farrell and Gibbons (1989), Newman and Sansing (1993) and Goltsman and
Pavlov (2011)). However, much of this literature has restricted attention to cheap talk
communication. In contrast, the sender engages in different types of communication in
3
our model: public disclosures are verifiable but internal communication is via cheap talk.
Moreover, the incentives for the two types of communication are interdependent due to
leakage. The manager’s incentive to distort her cheap-talk message to employees arises
endogenously because her internal communication can be leaked publicly, and such leaks
affect the short-term stock price.
The rest of the paper is as follows. Section 2 introduces the model. Section 3 charac-
terizes the equilibrium disclosure and internal communication by the manager, and Section
4 presents the comparative statics and efficiency results for the benchmark model with no
conflicts of interest. Section 5 extends the analysis to the case where the manager’s incen-
tives are misaligned with the other stakeholders. Section 6 discusses some implications of
the model and concludes. Unless noted otherwise, proofs are in the Appendix.
2 Model
There are three dates t ∈ {1, 2, 3}, and three types of players: a manager M , a continuum of
shareholders s ∈ S, and a continuum of employees i ∈ I. The value of the firm, V , depends
on its fundamentals θ, and the action xi of its employees. We assume that the value is
increasing in fundamentals and increasing in the degree of coordination across its employees.
Specifically, suppose the firm’s value is given by:
V = θ − β∫i∈I
(xi − θ)2di, (1)
where β > 0 measures the relative importance of coordination within the firm. For tractabil-
ity, suppose that θ is uniformly distributed between l and h > l i.e., f (θ) = 1h−l .
2
Date t = 1. At date t = 1, the manager observes fundamentals θ with probability π,
and denote the indicator variable for this event by ξ ∈ {0, 1} (so that Pr (ξ = 1) = π).3
An informed manager chooses whether to publicly disclose her information (i.e., d = θ) or
not (i.e., d = ∅). This public disclosure is observable by both shareholders and employees.
Moreover, while the manager can choose not to publicly disclose her information, we assume
she cannot publicly lie, i.e., such external disclosures are verifiable. The manager can also
internally communicate to her employees via a cheap-talk message m. This internal com-
munication is not verifiable (i.e., the manager can lie) and not immediately observable to
2The linear-quadratic specification for V and the distributional assumption for θ are made for tractability.We expect the qualitative nature of our results to survive in more general settings, but a formal analysis isnot immediately tractable.
3Since π < 1, voluntary disclosure is not fully unraveling. This approach has been extensively adopted inthe literature starting with Dye (1985) and Jung and Kwon (1988).
4
shareholders.
Date t = 2. At date t = 2, employees optimally choose their action to maximize the
value of the firm. In particular, employee i chooses action xi to maximize:
xi ≡ arg maxx
E[V∣∣m, d] (2)
⇒ xi = E[θ∣∣m, d] . (3)
Importantly, the message m to employees may be leaked to shareholders with probability
ρ, and denote the indicator variable for this event by λ ∈ {0, 1}. Given their information,
shareholders set the price P of the firm equal to its conditional expectation i.e.,
P = E[V∣∣d, λ×m× ξ] . (4)
Date t = 3. At the final date, the value of the firm is publicly revealed. The manager
receives a payoff U which depends on a weighted average of the date 2 price and the final
value V of the firm:
U = (1− δ)P + δV, (5)
where δ ∈ [0, 1). We assume that δ < 1 to ensure that the manager’s payoff depends on the
price P , and so her disclosure policy is non-trivial. Specifically, when δ = 1, the manager
can only affect her payoff through the information she conveys to her employees. As a result,
she is indifferent between a large class of disclosure / messaging strategies, as long as they
ensure employees are perfectly informed about θ.4
Moreover, to ensure that an informed manager has an incentive to withhold information
for sufficiently low fundamentals, we make the following assumption.
Assumption 1. Assume that h+l2− β (h−l)2
12> l.
When the manager is uninformed about fundamentals, the unconditional expected value
of the firm is given by
E [V ] = E [θ]− βE[(θ − E [θ])2] = h+l
2− β
12(h− l)2 . (6)
In contrast, by revealing her information perfectly to her employees (and thereby enabling
perfect coordination), a manager of type θ = l can ensure that the firm has value of V = l.
Assumption 1 ensures that a manager of the lowest type (i.e., θ = l) would strictly prefer to
withhold her type and pool with the uninformed managers.
4For instance, she is indifferent between disclosing θ publicly, or not disclosing θ to shareholders, butdisclosing it perfectly to shareholders.
5
We focus on pure strategy, Perfect Bayesian equilibrium. In particular, an equilibrium
is characterized by: (i) communication strategies {d, m} that maximize the manager’s ex-
pected utility U at date 1, (ii) optimal employee actions xi that maximize the conditional
expected value of the firm, (iii) a pricing rule given by (4), and (iv) participants’ beliefs
that satisfy Bayes’ rule wherever it is well-defined. Since the manager communicates with
her employees using cheap talk, there always exist equilibria that feature babbling. We shall
instead focus on characterizing equilibria that feature informative cheap-talk communication
between the manager and employees.
Remark 1. The manager’s payoff specification in (5) implies that there is no explicit conflict
of interest between manager and the shareholders or employees. This serves as a useful
benchmark in which to study the effects of leakage on the manager’s external and internal
communication. In Section 5, we relax this assumption by allowing for biases in the manager’s
payoffs that explicitly distort her incentives relative to those of the other stakeholders in the
firm.
Remark 2. In practice, leaks often reveal negative information about the firm. Our model
captures this feature, since in equilibrium, the manager withholds relatively negative infor-
mation, and this is the information that may be leaked to the market. However, a limitation
of our model is that the probability of leakage ρ is constant, and does not depend on either
the fundamentals θ of the firm, or the message m sent by the manager. This is primarily
for tractability. As a first step to exploring the robustness of our results to this restriction,
we consider an extension in Appendix B where the probability of leakage ρ is a decreasing
function of θ. While this assumption reflects the notion that worse news about fundamentals
is more likely to be leaked, it significantly limits our ability to characterize the equilibrium.
However, our numerical analysis suggests that the main conclusions of our benchmark analy-
sis persist: increasing the likelihood of leaks improves public disclosure but worsens internal
communication. We hope to extend this analysis and explore an endogenous model of what
information is leaked in future work.
3 Equilibrium Communication and Leakage
In this section, we characterize the equilibrium communication by the manager. We begin
with the benchmark case where there is no possibility of leakage (i.e., ρ = 0), and show that
in this case, an informed manager optimally chooses to perfectly reveal her information to
her employees. We then consider the case with leakage (i.e., ρ > 0). The possibility that
her internal communication is leaked to shareholders creates an endogenous incentive for the
manager to mislead employees — all else equal, she would prefer to report that fundamentals
6
are higher than they actually are. In equilibrium, this leads to noisy internal communication:
the manager’s cheap-talk message is only partially informative.
3.1 Preliminaries
We first characterize an uninformed manager’s strategy. Since external disclosures to share-
holders are verifiable, an uninformed manager must publicly report that she is uninformed
i.e., d = ∅. Moreover, conditional on her beliefs, the value of the firm is maximized when
maxxi
E [V (θ, {xi}i)] = maxxi
E [θ]− βE[∫
i
(θ − xi)2 di
](7)
⇒ x∗i = E [θ] = h+l2. (8)
But the manager can induce this action by revealing to her employees that she is uninformed
i.e., m = {θ ∈ [l, h]}. This gives us the following result.
Lemma 1. In any equilibrium, an uninformed manager discloses that she is uninformed to
shareholders and employees i.e., d = ∅ and m = {θ ∈ [l, h]}.
The above result implies that in the following analysis, we can characterize the equilib-
rium by specifying the strategy of the informed manager, since an uninformed manager has
the same communication strategy in equilibrium.
Next, consider an informed manager who observes a realization θ of fundamentals. Since
external disclosures are verifiable and observable by both shareholders and employees, a
disclosure of θ implies that P = θ and
V = θ − β∫i∈I
(E [θ|θ]− θ)2di = θ, (9)
so that the manager’s payoff from disclosing θ is UD (θ) = θ, irrespective of whether or not
there is a leak. Moreover, this implies that the manager’s optimal disclosure policy must
follow a cutoff rule i.e., she should only disclosure θ when θ ≥ q for some disclosure threshold
q. This implies the following observation.
Lemma 2. In any equilibrium, an informed manager’s optimal disclosure decision follows
a cutoff strategy.
Next, we turn to the benchmark case without leakage.
7
3.2 Equilibrium with no leakage
Suppose there is no possibility of leakage i.e., ρ = 0. Since the message to her employees
will not be revealed to the shareholders, an informed manager has no incentive to distort
her internal communication. In this case, we have the following result.
Proposition 1. Suppose Assumption 1 holds and there is no leakage i.e, ρ = 0. Let
qNoLeak = h− (h− l) 6−√
6(1−π)(6−πβ(h−l))6π
. (10)
Then, there exists an equilibrium in which an informed manager (i) discloses θ to sharehold-
ers if and only if θ ≥ qNoLeak, and (ii) perfectly reveals θ to her employees.
In the equilibrium described above, the manager perfectly reveals her information to her
employees, and so the value of the firm (conditional on the manager being informed) is θ.
The overall payoff to the manager depends on whether she discloses θ to shareholders. If she
does, then her payoff is given by
UD (θ) = (1− δ)P + δθ = θ, (11)
since P = E [V |θ] = θ. On the other hand, by not disclosing her information publicly, she
can pool with the uninformed managers, which yields a price
PND(qNoLeak
)=
(1− π)E[V∣∣ξ = 0
]+ π Pr
(θ < qNoLeak
)E[V∣∣ξ = 1, d = ∅
]1− π + π Pr (θ < qNoLeak)
(12)
and an overall payoff of
UND (θ) = (1− δ)PND + δθ. (13)
The equilibrium disclosure cutoff qNoLeak is pinned down by the natural indifference condi-
tion: a manager with fundamentals θ = qNoLeak is indifferent between disclosing her infor-
mation to shareholders and not i.e.,
UND(qNoLeak
)= UD
(qNoLeak
). (14)
As we show in the proof of Proposition 1, Assumption 1 ensures the existence of such a cutoff
qNoLeak ∈ [l, h].
The cutoff threshold qNoLeak is decreasing in β and π. All else equal, a higher β reduces
the value of an “uninformed” firm (i.e., the expected value of the firm conditional on the
manager being uninformed), and consequently decreases the benefit of pooling with the
8
uninformed managers. This leads an informed manager to disclose more often. Similarly,
when π is higher, the benefit from not disclosing θ is lower since conditional on no disclosure,
shareholders attribute a higher likelihood of the manager being informed. In the limit, when
π = 1, shareholders assume that the manager is informed irrespective of whether she discloses
or not, and as a result, there is no benefit from withholding information i.e., qNoLeak = l.
Proposition 1 characterizes the equilibrium with the most informative internal commu-
nication. Since the manager uses cheap talk to communicate with her employees, there exist
other equilibria in which internal communication is coarser. However the above result es-
tablishes that, in the absence of leaks, perfectly informative cheap talk can be sustained in
equilibrium. Next, we show that this is no longer the case when there is leakage.
3.3 Equilibrium with leakage
The possibility that messages to her employees may be leaked to shareholders distorts an
informed manager’s external and internal communication decisions. On the one hand, the
likelihood of a leak reduces the manager’s incentives from withholding information from
shareholders. On the other hand, the possibility of a leakage creates an incentive for the
manager to distort her message to employees: because the message will be leaked to share-
holders with positive probability, she has an incentive to bias it upwards relative to the true
θ. This incentive makes it impossible to sustain fully informative cheap talk between the
manager and her employees. However, as the following result establishes, there may exist
equilibria with partially informative cheap talk.
Proposition 2. Suppose Assumption 1 holds and there is leakage i.e, ρ > 0. Then there
exists a positive integer N ≥ 1 such that, for every N with 1 ≤ N ≤ N , there exists an
equilibrium characterized by cutoffs l = a0 < ... < aN = qLeakN ≤ h, where an informed
manager (i) discloses θ to shareholders if and only if θ ≥ qLeakN , (ii) perfectly reveals θ to her
employees if and only if θ ≥ qLeakN , and (iii) reports only the partition θ ∈ [an, an+1] to her
employees when θ < qLeakN . There always exists an equilibrium for N = 1. Moreover, for a
fixed N ≤ N , the disclosure cutoff qLeakN is unique.
We leave the details of the proof to the appendix, but provide an outline of the steps here.
As in the case with no leakage, note that conditional on disclosure, the value of the firm and
its price are given by θ, since both shareholders and employees can observe fundamentals
perfectly. In this case, the payoff to the manager is UD (θ) = θ.
In contrast to the case with no leakage, however, when the manager does not disclose her
signal to shareholders, her communication to employees is also coarse. Instead of revealing θ
9
to her employees, the manager only reports the partition [an, an+1] that it is in. This implies
that, conditional on no disclosure, the value of the firm is given by
VI (θ) = θ − β(an+an+1
2− θ)2, (15)
which reflects the lack of coordination within the firm due to less informative internal com-
munication.
The price of the firm depends on whether or not a leak happens. Conditional on a leak,
the price is given by the conditional expectation of the value of the firm, given that θ is in
the reported partition:
PI,L = E[VI (θ)
∣∣θ ∈ [an, an+1]]. (16)
Conditional on no leakage, the price reflects a weighted average of two possible cases: (i) the
manager is uninformed (i.e., ξ = 0) or (ii) the manager is informed, did not disclose and the
information was not leaked (i.e., ξ = 1, d = ∅ and λ = 0). As a result, the price conditional
on no leakage is given by
PNL =(1− π)E
[V∣∣ξ = 0
]+ π (1− ρ) Pr
(θ < qLeakN
)E[V∣∣ξ = 1, d = ∅, λ = 0
]1− π + π (1− ρ) Pr
(θ < qLeakN
) . (17)
By not disclosing her information, the expected price that the manager gets is therefore
given by:
PND = ρPI,L + (1− ρ)PNL, (18)
and so her expected payoff from not disclosing is:
UND (θ, [an, an+1]) = (1− δ)PND + δ(θ − β
(an+an+1
2− θ)2). (19)
Finally, we can use the manager’s indifference between disclosure and no disclosure at θ =
qLeakN to characterize the disclosure cutoff i.e.,
UD(qLeakN
)= UND
(qLeakN ,
[aN−1, q
LeakN
]), (20)
and indifference between adjacent messages {θ ∈ [an−1, an]} and θ ∈ [an, an+1] at θ = an i.e.,
UND (an, [an−1, an]) = UND (an, [an, an+1]) , (21)
to characterize the sequence of cheap-talk cutoffs {an}Nn=0. The sufficient condition for the
existence of these cutoffs produces an upper bound on the informativeness of the internal
10
communication.
Proposition 2 highlights the main channel through which leaks affects the manager’s
communication strategy. Because her internal communication to employees is leaked to
shareholders with positive probability, the manager has an incentive to distort her message
in an effort to push the price up. This misalignment in incentives makes fully informative
cheap-talk impossible, and as a result, internal communication in equilibrium is coarse. As
we discuss in the next section, this can lead to a decrease in real efficiency.
4 Comparative statics and real efficiency
In this section, we first characterize how the informativeness of internal and external com-
munication changes with the parameters of the model. This analysis reveals a key tradeoff
between internal and external communication in our setting. An increase in the probability
of leaks decreases the cost of public disclosure, but also makes informative internal commu-
nication more difficult to sustain. As a result, when leaks are more likely, public disclosure is
higher, but internal communication is less informative. We then describe how the possibility
of leaks affect real efficiency. In our setting, the source of inefficiency is the lack of coordi-
nation by employees. We show that real efficiency is maximized either when there are no
leaks, or when leaks happen with probability one. However, for intermediate probabilities,
real efficiency is lower than at either extreme.
4.1 Comparative statics
As Proposition 2 highlights, there exist multiple equilibria, each with a different level of in-
formativeness of internal communication (i.e., a different N). Moreover, the level of external
disclosure (i.e., the cutoff qLeakN ) itself depends on the informativeness of internal communica-
tion (i.e., N). To develop our intuition for how these equilibrium measures of informativeness
depend on the underlying parameters, we first characterize how qLeakN changes as a function
of parameters, treating N as a (fixed) parameter. We then study how N , the upper bound
on N that characterizes the equilibrium with the most informative internal communication,
changes as a function of parameters. We conclude with a set of numerical illustrations of
the overall effects on the cutoff qLeakN that highlight the interaction of these two effects.
The next result characterizes how qLeakN changes with parameters for a fixed N .
Proposition 3. Suppose there is leakage (i.e., ρ > 0). For a fixed N < N , the disclosure cut-
off decreases in the probability of a leak (ρ), the probability that the manager is informed (π),
and the value relevance of coordination (β). Moreover, holding other parameters fixed, the
11
disclosure cutoff qLeakN decreases as the informativeness of internal communication decreases
(i.e., N decreases).
The above result characterizes the impact of leaks on public disclosure. The manager’s
incentives to communicate are driven by two considerations: she would like to strategically
withhold negative information from shareholders in order to increase the price P , but she
would prefer to be as informative to employees as possible to encourage coordination. The
presence of leaks affects both these channels. Leaks decrease the benefits of withholding
information, since there is a positive probability this information is made public anyways.
Leaks also decrease the informativeness of internal communication since they create incen-
tives for the manager to distort her message to employees. On the margin, this encourages
the manager to disclosure more information publicly in an effort to mitigate the effect of
noisier internal communication.
Fixing the informativeness of internal communication, this implies that an increase in the
probability of a leak increases public disclosure by the manager (decreases qLeak) since it is
less beneficial to withhold information. Moreover, for a fixed probability of leaks, when the
informativeness of internal communication is lower (i.e., N decreases), the manager partially
compensates the decrease in coordination by increasing disclosure. As a result, the disclosure
threshold qLeakN decreases as N decreases.
The effects of π and β on the disclosure threshold are more direct. An increase in the ex-
ante likelihood of being informed, π, decreases the incentives for informed managers to pool
with the uninformed managers by withholding information, and as a result, leads to greater
disclosure. For a fixed level of informativeness of internal communication, an increase in the
relative importance of coordination β encourages the manager to disclose more information
publicly.
Next, we characterize the impact of leaks on the maximal informativeness of internal
communication.
Proposition 4. The maximum informativeness of internal communication that can be sus-
tained (i.e., N) weakly decreases with the probability of a leak (ρ) and the probability that
the manager is informed (π).
Note that while an increase in N leads to an increase in the corresponding threshold
qLeakN , the maximal informativeness N decreases when the manager is more likely to disclose
information. Intuitively, N measures how finely the range of undisclosed θ can be partitioned.
When there is more disclosure, the range of the undisclosed θ is smaller, and consequently,
N is lower.
12
Figure 1: Disclosure versus leakage
The figure plots the disclosure threshold q, when (i) there are no leaks and (ii) when thereare leaks for different values of N . Benchmark parameters are h = 2, l = 0, δ = 0.75, β = 1,and π = 0.5.
0.0
0.2
0.4
0.6
0.00 0.25 0.50 0.75 1.00ρ
q
No LeakLeak, N = 1Leak, N = 2Leak, N = 3Leak, N = 4
This observations helps explain the intuition for how π and ρ affect N . When the prior
probability that the manager is informed (π) increases, the benefit from withholding infor-
mation decreases. All else equal, the manager is more likely to disclosure θ, but this implies
the non-disclosure region is smaller, and as a result, N is lower. Similarly, increasing the
probability of a leak (ρ) decreases the manager’s cost of public disclosure, but this makes
it more difficult to sustain a higher level of N . Intuitively, a higher ρ increases the man-
ager’s incentives to distort internal communication, since it is more likely to be leaked to
shareholders and, thereby, affect the price.
Figure 1 provides a numerical illustration of the overall effect of leakage on the disclosure
threshold. First, note that disclosure is always lower without leakage, as evidenced by the
solid line in the figure. Second, consistent with Proposition 3, when there is leakage, the
cutoff qLeakN increases with N but decreases with ρ for a fixed N . Moreover, the maximal
level of informativeness, N , decreases with ρ as implied by Proposition 4, since informative
communication is more difficult to sustain with higher leakage. As such, the effects on qLeakN
and N reinforce each other: an increase in ρ decreases qLeakN for a fixed N , and decreases N ,
which leads the disclosure threshold in the most internally informative equilibrium to fall.
13
4.2 The effect of leaks on efficiency
The source of inefficiency in our setting is the lack of coordination by the employees. Specif-
ically, real efficiency is given by
RE (ρ,N) = −E[β
∫i∈I
(x∗i − θ)2
](22)
=
0 with no leakage
−βF(qLeakN
)E[∫
i∈I (x∗i − θ)2∣∣θ < qLeakN , {an}Nn=1
]with leakage
. (23)
This implies the following result.
Proposition 5. Real efficiency is maximized when either (i) there is no leakage (i.e., ρ = 0),
or (ii) leaks occur with probability 1 (i.e., ρ = 1).
On the one hand, when there is no leakage, Proposition 1 implies that the manager
perfectly reveals her information to employees, and as a result, coordination is perfect. On
the other hand, as we show in the proof of the above result, when leaks occur with probability
1 (i.e., when ρ = 1), there is no benefit from withholding information from shareholders,
and so the manager discloses all her information (i.e., disclosure threshold is qLeakN = l).5
As a result, employees observe fundamentals perfectly, and again, coordination is perfect.
When the probability of leaks is positive, but less than one, real efficiency is lower than
in the extremes since neither public disclosures nor internal communication are perfectly
informative. As a result, employees only have noisy information about fundamentals and
the firm incurs an efficiency loss due to decreased coordination.
Figure 2 illustrates how real efficiency changes with the probability of leakage. The overall
effect of ρ on real efficiency is non-monotonic, and depends on the level of internal commu-
nication (i.e., N). This relation is driven by two effects. First, consistent with Proposition
4, informative internal communication is more difficult to sustain as ρ increases i.e., N falls
with ρ. Second, for a fixed N , efficiency tends to decrease with ρ when informativeness is
high (e.g., N = 3, 4), but increases with ρ when informativeness is low (e.g., N = 1).
To better understand the relation between efficiency and ρ for a fixed N , note that an
increase in ρ has two offsetting effects. On the one hand, an increase in the probability of
leaks (higher ρ) increases public disclosure (decreases q), which improves efficiency. On the
other hand, more leakage increases the manager’s incentives to misreport her information
internally and, consequently, makes internal communication less informative. The first effect
5Arguably, leakage with probability one is an unrealistic benchmark in general, since there are other(unmodeled) constraints on what information can be revealed publicly.
14
Figure 2: Real efficiency versus leakage
The figure plots real efficiency RE as a function of parameters. Benchmark parameters areh = 2, l = 0, δ = 0.75, β = 1, and π = 0.5.
−0.003
−0.002
−0.001
0.000
0.00 0.25 0.50 0.75 1.00ρ
RE
No LeakLeak, N = 1Leak, N = 2Leak, N = 3Leak, N = 4
dominates when internal communication is not very informative (i.e., N is low), but the
second effect dominates when internal communication is informative (e.g., N is high). As
a result, real efficiency initially decreases but eventually increases with ρ and reaches its
maximum at ρ = 1, when public disclosure is perfectly informative. In general, the analysis
suggests that leaks are more likely to improve efficiency when internal communication is less
informative. This observation has potentially important implications for regulatory policy,
which we return to in our concluding remarks.
The results of this section imply that leaks have a negative effect: the possibility of
leakage tends to decrease real efficiency, even though it can lead to more public disclosure.
The next section considers a setting in which the possibility of leaks can play a more positive
role.
5 The effect of leaks with managerial bias
Leaks are often ascribed to have a disciplinary role: the possibility of leaks discourages
managers from misbehaving because they are concerned about the consequences of their
behavior becoming public. Our benchmark model does not capture this aspect, since the
incentives of the manager are well aligned with those of other stakeholders of the firm. As a
result, leaks are counterproductive because they limit the manager’s ability to communicate
with her employees.
However, we can analyze the disciplinary role of leaks by introducing a conflict of interest
15
between the manager and the firm’s other stakeholders. In particular, suppose the value of
the firm is given by (1), the employee’s optimal action is given by (3), and the shareholders’
pricing rule is given by (4). However, now assume the manager’s payoff is given by
U = (1− δ)P + δV , where (24)
V = V (θ + b) = θ + b− β∫i∈I
(xi − (θ + b))2 di, (25)
for b ≥ 0.6 The parameter b parsimoniously captures the degree to which the manager’s
incentives differ from those of the employees. This bias may arise because the manager has
different priors over the firm’s fundamentals (e.g., she is more optimistic about the firm so
that b > 0), or if she prefers coordinating employees’ actions towards a different objective.
The additive specification is a standard approach to introducing such conflicts of interest,
and is chosen primarily for tractability (e.g., Crawford and Sobel, 1982).
In this case, the equilibrium is characterized by the following result.
Proposition 6. Suppose Assumption (1) holds and b > 0. Then there exists a positive inte-
ger N ≥ 1 such that for every N with 1 ≤ N ≤ N , there exists an equilibrium characterized
by cutoffs l = a0 < ... < aN = qBiasN ≤ h, where an informed manager (i) discloses θ to
shareholders if and only if θ ≥ qBiasN , (ii) perfectly reveals θ to her employees if and only if
θ ≥ qBiasN , and (iii) reports only the partition θ ∈ [an, an+1] to her employees when θ < qBiasN .
There always exists an equilibrium for N = 1. Moreover, for a fixed N ≤ N , the disclosure
cutoff qBiasN is unique.
The proof of this result extends that of Proposition 2 by accounting for the bias in the
manager’s payoff. The key observation is that even when there is no leakage (i.e., ρ = 0),
the manager has an incentive to distort her report to employees, and consequently, internal
communication is not perfectly informative.
When the manager publicly discloses her signal, both employees and shareholders observe
θ. In this case, P = θ, and so the manager’s payoff is
UD (θ) = (1− δ) θ + δ(θ + b− βb2
). (27)
Conditional on no disclosure, the manager reports that θ ∈ [an, an+1] to her employees and
6The proof of the results in this section allow for a slightly more general specification for V , given by:
V = θ + b0 − β∫i∈I
(xi − (θ + b))2di (26)
where b and b0 can potentially be different. The key parameter is the bias in the quadratic term b since thisaffects the manager’s cheap talk strategy.
16
her resulting payoff is
UND (θ, [an, an+1]) = (1− δ)PND + δ(θ + b− β
(an+an+1
2− θ − b
)2). (28)
Note that these expressions mirror the corresponding expressions for the benchmark model
in Section 3.3, but account for the bias terms b0 and b. As before, the disclosure cutoff is
pinned down by the indifference between disclosure and not at θ = qBiasN i.e.,
UD(qBiasN
)= UND
(qBiasN ,
[aN−1, q
BiasN
]), (29)
while the indifference condition between adjacent messages {θ ∈ [an−1, an]} and {θ ∈ [an, an+1]}at θ = an i.e.,
UND (an, [an−1, an]) = UND (an, [an, an+1]) , (30)
pins down the sequence of cheap-talk cutoffs {an}Nn=0. An immediate consequence of the
above indifference conditions is that the equilibrium disclosure and communication policies
does not depend on b0.
Intuitively, when the manager’s incentives are less aligned with those of the employees
(i.e., when b is larger), internal communication is less informative, even in the absence of
leakage. As a result, N decreases with b. Moreover, this decrease in informativeness decreases
the value of the firm due to worse coordination, which in turn, increases the relative benefit
from publicly disclosing information. As a result, for a fixed N , the disclosure cutoff qBiasN
falls with the bias b i.e., a larger bias leads to more disclosure. These results are summarized
in the next result.
Proposition 7. Suppose N ≤ N and the manager is biased. For a fixed N , the disclosure
cutoff qBiasN decreases in the bias b. Moreover, the upper bound on the informativeness of
internal communication (i.e., N) decreases in the bias b.
Panel (a) of Figure 3 provides an illustration of the above results. The figure plots the
disclosure cutoffs as a function of the bias b when there are no leaks (solid lines) and when
there are leaks (ρ = 0.25, dotted lines). The plots suggest that in either case: (i) the cutoffs
are decreasing in b, and (ii) the maximal informativeness N is decreasing in b. Moreover, as
in the benchmark base (with a bias b = 0), the disclosure cutoff is lower in the presence of
leaks (i.e, dotted lines are lower than solid lines).
Panel (b) of Figure 3 also confirms that the effect of ρ on the disclosure cutoff and the
informativeness of internal communication are robust to the introduction of the managerial
bias b. Specifically, an increase in the probability of leakage (i.e., ρ) leads to a decrease in the
maximal informativeness of internal communication (i.e., N) and, for a fixed N , a decrease
17
Figure 3: Disclosure cutoffs with a biased manager
The figure plots the disclosure cutoff qBiasN as a function of parameters. Benchmarkparameters are h = 2, l = 0, δ = 0.75, β = 1, π = 0.5, ρ = 0.25 and b = 0.01.
0.35
0.40
0.45
0.50
0.000 0.025 0.050 0.075 0.100b
q
No Leak, N = 1No Leak, N = 2Leak, N = 1Leak, N = 2
0.0
0.2
0.4
0.00 0.25 0.50 0.75 1.00ρ
q
No Leak, N = 1No Leak, N = 2No Leak, N = 3Leak, N = 1Leak, N = 2Leak, N = 3
(a) qBiasN vs. b, with ρ = 0.25 (b) qBiasN vs. ρ with b = 0.01
in the disclosure threshold qBiasN . As before, leaks make internal communication coarser, but
increase public disclosure.
Next, we turn to real efficiency when the manager is biased. When the manager’s bias is
b, the probability of leakage is ρ and the number of partitions of the internal communication
is N , denote real efficiency by RE (ρ,N, b), where
RE (ρ,N, b) = −E[β
∫i∈I
(x∗i − θ)2
]= −βF
(qBiasN
)E[∫
i∈I(x∗i − θ)
2∣∣θ < qBiasN , {an}Nn=1
](31)
Since internal communication is noisy when the manager is biased, real efficiency is no longer
maximized when there is no leakage (i.e., ρ = 0) as in the benchmark case. Moreover, as
the following result highlights, real efficiency may be higher with some leakage than with no
leakage.
Proposition 8. Real efficiency is maximized when leaks occur with probability 1 (i.e., ρ =
1). Moreover, for a fixed b > 0, there exists a threshold ρ > 0, such that for ρ ≥ ρ,
RE (ρ,N = 1, b) > RE(0, N = N , b
).
In contrast to our benchmark model, the above result implies that leaks can have a
disclipinary role when there are conflicts of interest between the manager and the employees.
The bias in the manager’s payoff induces coarse internal communication even in the absence
of the leaks. While increasing the probability of a leak decreases the informativeness of
internal communication further, it increases public disclosure. If internal communication
is sufficiently informative initially (i.e., N is high), then the first effect dominates, and an
18
Figure 4: Real efficiency with a biased manager
The figure plots real efficiency RE as a function of parameters. Benchmark parameters areh = 2, l = 0, δ = 0.75, β = 1, π = 0.5,ρ = 0.25 and b = 0.01.
−0.0035
−0.0030
−0.0025
−0.0020
−0.0015
0.000 0.025 0.050 0.075 0.100b
RE
No Leak, N = 1No Leak, N = 2Leak, N = 1Leak, N = 2
−0.003
−0.002
−0.001
0.000
0.00 0.25 0.50 0.75 1.00ρ
RE
No Leak, N = 1No Leak, N = 2No Leak, N = 3Leak, N = 1Leak, N = 2Leak, N = 3
(a) RE vs. b, with ρ = 0.25 (b) RE vs. ρ with b = 0.01
increase in ρ decreases efficiency. However, when the manager’s bias (or the probability of
leaks) is sufficiently high, informative internal communication can no longer be sustained
(i.e., N = 1). In this case, the second effect dominates and, as a result, an increase in ρ
increases efficiency.
Figure 4 illustrates these effects for a specific parameterization. Panel (a) compares
real efficiency for varying levels of managerial bias when there is zero or positive (ρ =
0.25) probability of leaks. When more informative internal communication is possible (e.g.,
N = 2), real efficiency tends to be lower with leakage (blue, dashed line) than without
(blue, solid line), unless the manager’s bias is extremely small. In contrast, when internal
communication is not informative (i.e., N = 1), the equilibrium with leaks is more efficient
for all levels of b, since public disclosure is higher. Similarly, for a fixed bias b > 0, panel
(b) illustrates how real efficiency eventually increases in the likelihood of leaks. Intuitively,
a non-zero bias in the manager’s incentives limits the maximal informativeness of internal
communication (N = 3 in our numerical example). When the likelihood of leaks is sufficiently
high, informative internal communication cannot be sustained (i.e., N = 1), but in this case,
efficiency improves with ρ.
The discussion in Section 4.2 on the relationship between efficiency and ρ carries over to
the case of biased manager here: efficiency initially decreases with ρ, but eventually increases
until it reaches a maximum for ρ = 1. However, unlike the benchmark case with no bias
(i.e., b = 0), the equilibria without leaks (i.e., ρ = 0) do not dominate the ones with ρ < 1,
since internal communication is imperfect even in the absence of leakage. In fact, when
internal communication is not informative (i.e., N = 1), an increase in the probability of
19
leakage improves efficiency for all values of ρ. As in the case with no conflicts of interest
(i.e., with b = 0), the possibility of leaks is more likely to improve efficiency when internal
communication is less informative. We turn to the implications of this observation in our
final section.
6 Implications and Concluding Remarks
As our analysis highlights, leaks introduce an important tradeoff: more leakage makes public
disclosure more informative, but internal communication less informative. As a result, the
effect of leaks on real efficiency (welfare) is non-monotonic, and depends crucially on the
informativeness of internal communication within the firm. While identifying appropriate
empirical measures of the underlying parameters is extremely challenging, our model makes
a number of testable predictions.
First, the average price reaction to information leaks is negative. This is intuitive: an
informed manager publicly discloses good news, but withholds bad news (as in standard
disclosure models). Hence, any leaks reveal negative news to the market. This is consistent
with a number of papers in the literature, including Bowen et al. (2010) and Dyck et al.
(2010).
Second, firms respond to an increase (decrease) in the probability of leaks by increasing
(decreasing, respectively) public voluntary disclosures. For instance, our model predicts
that voluntary disclosures by firms should increase following the implementation of the
Whistleblower Program of the Dodd-Frank Act (see http://www.sec.gov/spotlight/dodd-
frank/whistleblower.shtml). However, a key challenge in testing such a prediction empiri-
cally is to separately identify the impact of such regulation on the likelihood of leakage, while
controlling for other aspects that might affect firms’ disclosure policies directly.
Third, all else equal, a higher likelihood of leakage is associated with less informative
internal communication. Empirical tests of this prediction are confounded by the fact that
following (the incidence of) a leak, firms usually make changes to the internal governance
and communication policies. As such a test of this prediction requires identifying firms that
are ex-ante more likely to have leaks, and comparing their internal communication to a
control group. Bowen et al. (2010) provide some preliminary evidence consistent with this
prediction: in their sample, targets of employee whistle-blowing allegations are more likely
to have unclear internal communications channels.
Fourth, an increase in the likelihood of leakage affects real efficiency across firms dif-
ferently. Specifically, real efficiency decreases in firms with fewer conflicts of interest and
informative internal communication, but can increase in firms with poor internal communi-
20
cation and bad governance. To the extent that firms with better governance and internal
communication perform better ex ante, our model predicts that the cross-sectional varia-
tion in firm performance should decrease in response to regulatory changes that encourage
leakage.
The above result is an important consideration when evaluating regulatory policies that
improve whistleblower protection. An increased likelihood of leakage can have a disciplinary
effect on firms, but not always. Such policies are more likely to be beneficial for firms that
have severe agency problems, uninformative internal communication and more short-term
compensation for managers (e.g., through stock / option grants). To the extent possible,
targeting such firms may be important in ensuring that regulations have the intended effect.
Our current model is stylized for tractability, but provides a first step in understanding
how external and internal communication is affected by leakage. Studying how leaks affect
the aggregation of information in a setting where employees have private information would
be a natural next step. It would also be interesting to develop a richer model of employee /
insider incentives to study the effect of strategic leakage and to endogenize what information
is leaked. We hope to explore these directions in future work.
21
References
Bowen, R. M., Call, A. C., Rajgopal, S., 2010. Whistle-blowing: Target firm characteristicsand economic consequences. The Accounting Review 85 (4), 1239–1271. 6
Crawford, V. P., Sobel, J., 1982. Strategic information transmission. Econometrica: Journalof the Econometric Society, 1431–1451. 1, 5
Dyck, A., Morse, A., Zingales, L., 2010. Who blows the whistle on corporate fraud? TheJournal of Finance 65 (6), 2213–2253. 6
Dye, R. A., 1985. Disclosure of nonproprietary information. Journal of Accounting Research,123–145. 1, 3
Farrell, J., Gibbons, R., 1989. Cheap talk with two audiences. The American EconomicReview 79 (5), 1214–1223. 1
Goltsman, M., Pavlov, G., 2011. How to talk to multiple audiences. Games and EconomicBehavior 72 (1), 100–122. 1
Jung, W.-O., Kwon, Y. K., 1988. Disclosure when the market is unsure of informationendowment of managers. Journal of Accounting Research, 146–153. 1, 3
Newman, P., Sansing, R., 1993. Disclosure policies with multiple users. Journal of AccountingResearch, 92–112. 1
22
Appendices
A Proofs of main results
Proof of Proposition 1. By Lemma 2, we know that the optimal disclosure strategy mustbe a cutoff strategy where the manager discloses θ only if θ ≥ q for some cutoff q. Since thereis no leakage, an informed manager optimally communicates θ perfectly to her employees.This implies that
V (ξ = 1) = θ, V (ξ = 0) = E [θ]− βvar [θ] . (32)
Conditional on disclosing her information to shareholders, this implies UD (θ) = θ, which isstrictly increasing in θ. Conditional on no disclosure, the price is then given by:
PND(q) =(1− π)V (ξ = 0) + πF (q)E[V (ξ = 1, θ)|θ < q]
(1− π) + πF (q)(33)
=(1− π)
(h+l
2− β (h−l)2
12
)+ π q−l
h−ll+q2
(1− π) + π q−lh−l
(34)
At θ = q, the manager should be indifferent between disclosing or not, and so:
q = δq + (1− δ)PND(q) (35)
↔ q =(1− π)
(h+l
2− β (h−l)2
12
)+ π q−l
h−ll+q2
(1− π) + π q−lh−l
(36)
↔ q =6πh−(h−l)(6±
√6(1−π)(6−πβ(h−l))6π
. (37)
To ensure that the cutoff q ∈ [l, h] we must select the larger root i.e.,
q = h− (h− l)6−√
6(1−π)(6−πβ(h−l))6π
≡ qNoLeak. (38)
Finally, note that when Assumption 1 holds, we have:
h+l2− β
12(h− l)2 > l⇒ 6− β(h− l) > 0, (39)
which ensures qNoLeak is well-defined. Moreover, the above establishes that qNoLeak is uniquelydetermined.
Proof of Proposition 2. Conditional on a message mn = {θ ∈ [an, an+1]}, note that
E [θ|mn] = an+an+1
2, var [θ|mn] = (an+1−an)2
12. (40)
Next, note that disclosure must follow a cutoff strategy (Lemma 2) and when θ is disclosedpublicly, the manager’s payoff from disclosure is UD (θ) = θ. When there is no disclosure,there are two possibilities:
23
• The manger is uninformed, in which case, the value of the firm is:
VNI = h+l2− β
12(h− l)2 = PNI . (41)
• The manager is informed, and only sends message mn to her employees. In this case,the value of the firm is
VI (θ,mn) = θ − β(an+an+1
2− θ)2. (42)
If the message is leaked to shareholders, the price is given by
PI,L (mn) = E[VI∣∣mn
]= an+1+an
2− β
12(an+1 − an)2 . (43)
On the other hand, if the message is not leaked, the price is given by
PI,NL (mn) = E[VI∣∣ξ = 1, d = ∅
]= E
[VI∣∣θ < q
]= qN+l
2− β
12
∑i
(an+1−an)3
qN−l. (44)
As a result, conditional on no disclosure, the expected price is then given by:
PND (mn) = ρPI,L + ρ (γPNI + (1− γ)PI,NL) , (45)
where ρ is the probability of a leak, and γ reflects the probability that the manager isuninformed, given that there is no disclosure (and no leak) i.e.,
γ ≡ 1− π1− π + π (1− ρ) Pr (θ < qN)
. (46)
As a result, an informed manager’s payoff from not disclosing her information is
UND (θ,mn) = δVI (θ,mn) + (1− δ)PND (mn) . (47)
The disclosure cutoff is pinned down by indifference for θ = q i.e.,
UD (q) = UND (q, {θ ∈ [aN−1, q]}) , (48)
while the cheap-talk partitions are characterized by indifference at the cutoffs θ = an i.e.,
UND (an, {θ ∈ [an−1, an]}) = UND (an, {θ ∈ [an, an+1]}) . (49)
The latter indifference condition implies that an satisfies the difference equation:
an−1−an+1
2
(ρ (1− δ)− β (3δ + (1− δ)ρ) (an−1+an+1−2an)
6
)= 0 (50)
⇒ an+1 ={an−1, 2an − an−1 + 6ρ(1−δ)
β(3δ+ρ(1−δ))
}(51)
Given the monotonicity of ai, the second solution solution must hold. The recursive equation
24
of the form:ai+2 = 2ai+1 − ai + 2c, a0 = l,
has the solution:an = l + A0n+ cn(n− 1)
for a constant A0 and
c ≡ 3(1− δ)ρβ(3δ + (1− δ)ρ)
. (52)
Since aN = q, we must have:
q − l = A0N + cN(N − 1) ≥ cN(N − 1). (53)
The constant A0 can be pinned down using aN = q, which implies:
A0 = q−lN− c(N − 1), (54)
⇒ an = (n−N)(nNc−l)+nqN
, aN−1 = l−(N−1)(cN−q)N
. (55)
Finally, note that the above implies:
Σ (q) ≡ 112(q−l)
N∑i=1
(ai − ai−1)3 =c2(N2−1)
12+ (q−l)2
12N2 . (56)
Given these expressions, one can solve for q (using (48)) as the fixed point to q = g (q),where
g (q) ≡ UND (q, {θ ∈ [aN−1, q]}) (57)
= δ(q − β
( q+aN−1
2− q)2)
+ (1− δ)
ρ( q+aN−1
2− β
12(q − aN−1)2)
+ (1− ρ) γ(h+l
2− β
12(h− l)2)
+ (1− ρ) (1− γ)(qN+l
2− βΣ (q)
) (58)
= q+l2− βΣ (q) + δ
(q−l2− 2βΣ (q)
)+ (1− δ)(1− ρ)γ
(h−q
2− β
((h−l)2
12− Σ (q)
))(59)
Fix an N ≥ 1 and recall that q − l ≥ cN (N − 1). To show the existence of a solution toq = g (q), where h > q > cN (N − 1) + l, it is sufficient to establish conditions for g (h) < hand g (l + cN (N − 1)) > l + cN (N − 1). When q = h, note that
g (h) = δVI + (1− δ) (ρPI,L + (1− ρ) (γPNI + (1− γ)PI,NL)) (60)
for VI , PI,L, PNI and PI,L<h, and so g (h) < h. Next, note that βc = 3(1−δ)ρ(3δ+(1−δ)ρ)
≤ 3, and so
g (cN (N − 1) + l) =
l + (1− δ) (1− ρ) γ(h−l
2− β (h−l)2
12
)+ cN(N−1)
2
(1 + δ − (1− δ) (1− ρ) γ
)−βc2((N−1)2+N2−1)
12
(1 + 2δ − (1− δ) (1− ρ) γ
) (61)
25
≥l + (1− δ) (1− ρ) γ
(h−l
2− β (h−l)2
12
)+ cN(N−1)
2
(1 + δ − (1− δ) (1− ρ) γ
)−3c(2N(N−1))
12
(1 + 2δ − (1− δ) (1− ρ) γ
) (62)
= l + (1− δ) (1− ρ) γ(h−l
2− β (h−l)2
12
)− cN(N−1)
2δ, (63)
where
γ =1− π
1− π + π (1− ρ) cN(N−1)h−l
. (64)
To ensure that g (cN (N − 1) + l) > l + cN (N − 1), we must have:
l +(1− δ) (1− ρ) (1− π)
1− π + π (1− ρ) cN(N−1)h−l
(h−l
2− β (h−l)2
12
)− cN(N−1)
2δ > l + cN (N − 1) (65)
(1− δ) (1− ρ) (1− π)(h−l
2− β (h−l)2
12
)1− π + π (1− ρ) cN(N−1)
h−l
> cN (N − 1)(1 + δ
2
)(66)
or equivalently,
cN (N − 1)(1 + δ
2
) (1− π + π(1−ρ)
h−l cN (N − 1))< (1− δ) (1− ρ) (1− π)
(h−l
2− β (h−l)2
12
).
(67)Since the LHS of the above is always increasing in N , the above condition is equivalent toensuring that N ≤ N for some bound. Moreover, note that the above condition holds forN = 1, and so N ≥ 1.
For a fixed N ≤ N , a sufficient condition for the uniqueness of qLeakN is that ∂g∂q< 1, since
this implies g (q) intersect the 45 degree line (at most once) from above. Note that
∂g
∂q=
12
(1 + δ − γ (1− δ) (1− ρ))− βΣq (1 + 2δ − (1− δ) (1− ρ) γ)
+γq (1− δ) (1− ρ)(h−q
2− β
((h−l)2
12− Σ
))(68)
whereΣq = 2(q−l)
12N2 > 0, γq = − π(1−π)(1−ρ)
(h−l)(1−π+π(1−ρ) q−lh−l)
2 < 0. (69)
At any point of intersection, we have q = g (q), and so
(1− δ) (1− ρ)(h−q
2− β
((h−l)2
12− Σ
))= 1
γ
(q −
(q+l2− βΣ (q) + δ
(q−l2− 2βΣ (q)
)))(70)
= 1γ
(q−l2
(1− δ) + (1 + 2δ) βΣ (q))≥ 0 (71)
Since 1 + 2δ − (1− δ) (1− ρ) γ > 0,
∂g
∂q< 1
2(1 + δ) < 1. (72)
This establishes that any point of intersection q = g (q), g (q) intersects the 45-degree
26
line from below, which in turn implies there can be only one such intersection for q ∈[l + cN (N − 1) , h].
Proof of Proposition 3. We begin by proving a useful Lemma.
Lemma 3. Fix an N < N . The unique solution qLeakN to q = g (q;N) satisfies:
qLeakN − l ≥ cN (N + 1) > cN (N − 1) . (73)
Proof. Uniqueness of qLeakN is established in the proof of Proposition 2. Specifically, recallthat for a fixed N , g (q) intersects the 45-degree line from above. Next, note that
G(q) ≡ g(q,N)− g(q,N − 1) = −(β(1 + 2δ − (1− δ)(1− ρ)γ)) [Σ(N)− Σ(N − 1)] (74)
where Σ(N)− Σ(N − 1) = 112
(c2(N2 − (N − 1)2) + (q − l)2 (N−1)2−N2
N2(N−1)2
)(75)
= 2N−112
(c2 − (q−l)2
N2(N−1)2
)(76)
This implies that for q = cN(N − 1), we have g(q,N) = g(q,N − 1). Since qLeakN − l ≥cN(N − 1), we have:
G(qLeakN ) = g(qLeakN , N)− g(qLeakN , N − 1) = qLeakN − g(qLeakN , N − 1) > 0 (77)
Moreover, since g(x;N) intersects the 45-degree line from above, we have:
l + cN(N − 1) ≤ g(l + cN(N − 1);N), (78)
and since G(l + cN(N − 1)) = 0, we have:
l + cN(N − 1) ≤ g(l + cN(N − 1);N − 1) (79)
But this implies that the solution qLeakN−1 to q = g(q;N − 1) must satisfy:
l + cN(N − 1) ≤ qLeakN−1 ≤ qLeakN (80)
which completes the result.Note that the above result immediately implies that qLeakN is increasing inN . Since q = g (q; ρ)and gq ≡ ∂g
∂q< 1, we have
∂q
∂ρ=
gρ1− gq
< 0 ⇔ gρ ≡∂g
∂ρ< 0. (81)
So to establish the comparative statics results for ρ, (and analogously π and β) it is sufficientto show gρ < 0 (and analogously gπ < 0 and gβ < 0). Given the expression for g (·) above,we have:
gρ = −Σρ (1 + 2δ − (1− δ)(1− ρ)γ) + (1− δ) ((1− ρ)γρ − γ)(h−q
2− β
((h−l)2
12− Σ
))< 0,
27
since γρ = γ(1−γ)1−ρ ≥ 0 (see (46)), h−q
2− β
((h−l)2
12− Σ
)> 0 (see the argument for (71) above)
and Σρ = ∂Σ∂c
∂c∂ρ> 0, given (52) and (56). Similarly, note that
gπ = (1− δ)(1− ρ)(h−q
2− β
((h−l)2
12− Σ
))γπ < 0 (82)
since γπ < 0 (see (46)) and h−q2− β
((h−l)2
12− Σ
)> 0 , and that
gβ = −(1 + 2δ − (1− δ)(1− ρ)γ)(Σ + βΣβ)− (1− δ)(1− ρ)γ (h−l)2
12, and (83)
Σ + βΣβ = 112
((qLeakN −l)
2
N2 − c2(N2 − 1)
)≥ 1
12
(c2(N + 1)2 − c2(N2 − 1)
)> 0, (84)
for N < N from Lemma 3, and so gβ < 0 for N < N .
Proof of Proposition 4. As seen in the proof of Proposition 3, for all N ≤ N , we have
l + cN(N − 1) ≤ g(l + cN(N − 1)), (85)
and for N we also havel + cN(N + 1) > g(l + cN(N + 1)). (86)
As π becomes larger, the left-hand side does not change for either inequality, while the right-hand side decreases since gπ < 0. For (86), this implies that the inequality holds for highervalues of π, and so N cannot increase with an increase in π. However, (85) may be violatedif we increase π sufficiently and N is reduced by one. This implies that N is non-increasingin π. Since cρ > 0 and gρ < 0, similar arguments across (85) and (86) imply that N isnon-increasing in ρ.
Proof of Proposition 5. Recall from the proof of Proposition 2 that qLeakN is the solutionto q = g (q), where g (q) is defined by
g (q) = q+l2− βΣ (q) + δ
(q−l2− 2βΣ (q)
)+ (1− δ)(1− ρ)γ
(h−q
2− β
((h−l)2
12− Σ (q)
)),
(87)
Σ (q) =c2(N2−1)
12+ (q−l)2
12N2 , c =3(1− δ)ρ
β(3δ + (1− δ)ρ), γ =
1− π1− π + π (1− ρ) q−l
h−l(88)
and
cN (N − 1)(1 + δ
2
) (1− π + π(1−ρ)
h−l cN (N − 1))≤ (1− δ) (1− ρ) (1− π)
(h−l
2− β (h−l)2
12
).
(89)When ρ = 1, the above condition reduces to:
cN (N − 1)(1 + δ
2
)(1− π) ≤ 0 ⇒ N = 1, (90)
28
i.e., when ρ = 1, N ≤ N = 1. Moreover, this implies that when ρ = 1,
Σ (q) = (q−l)2
12, c =
3(1− δ)β(3δ + (1− δ))
(91)
⇒ g (q)− q = − (q − l) + q−l2
(1 + δ)− β (q−l)2
12(1 + 2δ) (92)
= (q − l){−1 + 1+δ
2− β(1+2δ)
12(q − l)
}(93)
⇒ q =
{l, l − 6 (1− δ)
β (1 + 2δ)
}(94)
but since q ≥ l, the only possible solution is q = l. This implies that real efficiency ismaximized at ρ = 1 since there is full public disclosure.
Proof of Proposition 6. The proof follows the structure of the proof of Proposition 2. Inthe case there is disclosure, both insiders and outsiders observe θ. In this case,
PD = θ, UD = (1− δ) θ + δ(θ + b0 − βb2
)= θ + δ
(b0 − βb2
)(95)
Next, note that
V =
{h+l
2− β
12(h− l)2 + b0 − βb2 when ξ = 0
θ + b0 − β(an+1+an
2− θ − b
)2when ξ = 1
(96)
and conditional on a message mn = {θ ∈ [an, an+1]},
P =
h+l
2− β
12(h− l)2 when ξ = 0
an+1+an2− β
12(an+1 − an)2 when ξ = 1, k = 1 (leak)
q+l2− βΣ (q) when ξ = 1, k = 0 (no leak)
(97)
The expected price given no disclosure is:
PND (mn) =ρ(an+1+an
2− β
12(an+1 − an)2)
+ (1− ρ)(γ(h+l
2− β
12(h− l)2)+ (1− γ)
(q+l2− βΣ (q)
)) (98)
where as before, γ = 1−π1−π+π(1−ρ) q−l
h−l. Conditional on no disclosure, the manager’s payoff is
given by:
UND (θ,mn) = δ(θ + b0 − β
(an+1+an
2− θ − b
)2)
+ (1− δ)PND (m) (99)
The indifference condition to pin down qBiasN is given by:
UD (q) = UND (q, [aN−1, q]) (100)
⇔ q = δ(q − β
(( q−aN−1
2
)2+ 2b
( q−aN−1
2
)))+ (1− δ)PND ([aN−1, q]) (101)
29
The indifference conditions to pin down ai is given by
UND (an, [an, an+1])− UND (an, [an−1, an]) = 0 (102)
⇒−δβ
(an+1−an
2− b)2
+ (1− δ) ρ(an+1+an
2− β
12(an+1 − an)2)
−{−δβ
(an−1−an
2− b)2
+ (1− δ) ρ(an−1+an
2− β
12(an−1 − an)2)} = 0 (103)
Solving for an+1, we get
an+1 =
{an, 2an − an−1 +
6 ((1− δ) ρ+ 2βδb)
β (3δ + (1− δ) ρ)
}. (104)
Letting
c =3 ((1− δ) ρ+ 2βδb)
β (3δ + (1− δ) ρ), (105)
we have a solution of the type:
an = l + A0n+ cn (n− 1) . (106)
Since aN = q, we have: q − l = A0N + cN (N − 1) > cN (N − 1), or equivalently, N <
12
(√1 + 4(q−l)
c+ 1
). The expressions for aN−1 and Σ correspond to the no bias case with
the modified c. Note however, that
limρ→0
c =3 (2βδb)
β (3δ)= 2b 6= 0 (107)
and so even when there are no leaks, N < 12
(√1 + 2(q−l)
b+ 1
)i.e., internal communication
is coarse. Let
g (q) ≡ δ(q − β
(( q−aN−1
2
)2+ 2b
( q−aN−1
2
)))+ (1− δ)PND ([aN−1, q]) (108)
= g (q)− βbδ(cN (N − 1) + q − l
N
)(109)
where g (q) is given by (57), and note that qBiasN is the fixed point to q = g (q). To show theexistence of a solution to q = g (q), where h > q > cN (N − 1) + l, it is sufficient to establishconditions for g (h) < h and g (l + cN (N − 1)) > l + cN (N − 1). When q = h, note thatg (h) ≤ g (h) ≤ h. As before βc ≤ 3, and so analogous calculations to those in the proof ofProposition 2, we have:
g (l + cN (N − 1)) ≥ l + (1− δ) (1− ρ) γ(h−l
2− β (h−l)2
12
)− cN(N−1)
2δ − βbδ
(2cN(N−1)
N
),
(110)
30
where γ = 1−π1−π+π(1−ρ)
cN(N−1)h−l
. To ensure g (l + cN (N − 1)) ≥ l + cN (N − 1), we must have:
cN (N − 1)(1 + δ
2+ 2βbδ
N
) (1− π + π (1−ρ)cN(N−1)
h−l
)≤ (1− δ) (1− ρ) (1− π)
(h−l
2− β (h−l)2
12
).
(111)Note that this coincides with the corresponding condition in Proposition 2 when b = 0.Since the LHS is increasing in N , this corresponds to ensuring that N ≤ N for some bound.Moreover, the condition always holds for N = 1, which implies N ≥ 1. Finally, uniquenessof qBiasN requires ∂g
∂q< 1, which holds since (72) holds and
∂g
∂q=∂g
∂q− βbδ
N(112)
for b > 0.
Proof of Proposition 7. From the proof of Proposition 6, we know that qBiasN is the fixedpoint to q = g (q), where
g (q) = g (q)− βbδ(cN (N − 1) + q − l
N
)(113)
and g (q), which is given by (57), does not depend on b. This implies that for a fixed N ,
∂∂bqBiasN =
∂g∂b
1− ∂g∂q
≤ 0 (114)
since ∂g∂q< 1 and ∂g
∂b≤ 0. Next, note that the upper bound N is characterized by a strict
equality in condition (111). Let
L ≡ cN(N − 1
) (1 + δ
2+ 2βbδ
N
)(1− π + π
(1−ρ)cN(N−1)h−l
)(115)
and note that ∂L∂b
+ ∂L∂N
∂N∂b
= 0, or equivalently,
∂N∂b
= −∂L∂b∂L∂N
. (116)
Next, note that
∂L
∂b=
2βδ(N − 1)c(π(N−1)N(1−ρ)c
h−l − π + 1)
+cb
π(N−1)2N2(1−ρ)c( 2bβδN
+ δ2
+1)h−l
+(N − 1)N(
2bβδN
+ δ2
+ 1) (π(N−1)N(1−ρ)c
h−l − π + 1) (117)
31
and cb > 0. Moreover,
∂L
∂N=
c(2N − 1
) (1 + δ
2+ 2βbδ
N
)(1− π + π
(1−ρ)cN(N−1)h−l
)+π(1−ρ)
h−l cN(N − 1
) (1 + δ
2+ 2βbδ
N
)c(2N − 1
)−2βbδ
N2 cN(N − 1
)(1− π + π
(1−ρ)cN(N−1)h−l
) (118)
For N = 1, ∂N∂b
= ∂L∂b
= 0. For N > 1, we have ∂L∂b> 0 and
∂L
∂N=
c(2N − 1
) (1 + δ
2+ 2βbδ
N
)(1− π + π
(1−ρ)cN(N−1)h−l
)−2βbδ
Nc(N − 1
)(1− π + π
(1−ρ)cN(N−1)h−l
)+π(1−ρ)
h−l cN(N − 1
) (1 + δ
2+ 2βbδ
N
)c(2N − 1
) (119)
≥[(
2N − 1) (
1 + δ2
)+
2βbδ(2N−1)N
− 2βbδ(N−1)N
]c
(1− π + π
(1−ρ)cN(N−1)h−l
)(120)
≥ 0 (121)
which implies ∂N∂b< 0.
Proof of Proposition 8. First, note that when ρ = 1, the condition that determinesN (i.e., expression (111)) reduces to:
cN (N − 1)(1 + δ
2+ 2βbδ
N
)(1− π) ≤ 0⇒ N = 1. (122)
cN (N − 1)(1 + δ
2
)(1− π) ≤ 0 ⇒ N = 1, (123)
Moreover, this implies that when ρ = 1, Σ (q) = (q−l)2
12,
⇒ g (q)− q = q+l2− q − βΣ (q) + δ
(q−l2− 2βΣ (q)
)− βbδ (q − l) (124)
= − (q − l){
12− δ + βδb+ β
12(1 + 2δ) (q − l)
}(125)
⇒ q =
{l, l − 12bβδ + 6 (1− δ)
β(2δ + 1)
}(126)
but since q ≥ l, the only possible solution is qBiasN = l, which implies RE (ρ = 1) = 0.Fix a b > 0, and q0
N denote the cutoff when there is no leakage. Since c = 2b when ρ = 0,we have
Σ (ρ = 0, N) =(2b)2 (N2 − 1)
12+
(q0N − l)
2
12N2. (127)
In contrast, when there is leakage with probability ρ but N = 1, we have
Σ (ρ,N = 1) =(qρ1 − l)
2
12. (128)
32
Efficiency is higher with leakage when
q0N − lh− l
Σ (ρ = 0, N) >qρ1 − lh− l
Σ (ρ,N = 1) , (129)
or equivalently, (q0N − l
)( (2b)2(N2−1)12
+(q0N−l)
2
12N2
)>
(qρ1−l)2
12. (130)
Note that for a fixed b > 0, the LHS is strictly positive and bounded away from 0, sinceq0N > l and N ≤ N . On the other hand, one can show that q1 is always decreasing in ρ and
limρ→1 q = l. This implies there exists a ρ > 0 such that for all ρ > ρ, the above conditionholds.
B Type dependent probability of leakage
In this appendix, we consider an extension of the model in which the probability of leakageρ depends on the type of the firm θ. Specifically, suppose that the probability ρ is givenby a decreasing function of θ, i.e., ρ = ρ (θ). We assume that there is some leakage i.e.,ρ (θ) > 0 for some θ ∈ [l, h] and that ρ (θ) is decreasing in θ. The rest of the setup isas described in Section 2. As before, we consider equilibria characterized by a sequenceof cutoffs l = a0 < ... < aN = qLeakN ≤ h, where an informed manager (i) discloses θ toshareholders if and only if θ ≥ qLeakN , (ii) perfectly reveals θ to her employees if and only ifθ ≥ qLeakN , and (iii) reports only the partition θ ∈ [an, an+1] to her employees when θ < qLeakN .
Conditional on disclosure, the manager’s utility is given by UD (θ) = θ. Conditional onno disclosure,
• If the manager is not informed, the value of the firm is VNI = PNI = h+l2− β
12(h− l)2 .
• If the manager is informed, and only sends a message mn = {θ ∈ [an, an+1]} to heremployees, the value of the firm is
VI (θ,mn) = θ − β(an+1+an
2− θ)2. (131)
Moreover, if the message is leaked to shareholders, the price is given by
PI,L (mn) = an+1+an2− β
12(an+1 − an)2 , (132)
but if it is not leaked to shareholders, the price is given by
PI,NL (mn) = qN+l2− β
12
∑i
(an+1−an)3
qN−l. (133)
As a result, conditional on no disclosure, the expected price is given by:
PND (θ,mn) = ρ (θ)PI,L (mn) + (1− ρ (θ)) (γPNI + (1− γ)PI,NL) , where (134)
γ ≡ 1− π1− π + π (1− ρND) Pr (θ < qN)
, and (135)
33
ρND =1
qN − l
∫ qN
l
ρ (x) dx, (136)
and the expected utility to the manager from not disclosing is given by
UND (θ,mn) = δVI (θ,mn) + (1− δ)PND (θ,mn) . (137)
The disclosure cutoff is pinned down by indifference for θ = q i.e.,
UD (q) = UND (q, {θ ∈ [aN−1, q]}) , (138)
while the cheap-talk partitions are characterized by indifference at the cutoffs θ = an i.e.,
UND (an, {θ ∈ [an−1, an]}) = UND (an, {θ ∈ [an, an+1]}) . (139)
The latter condition implies that
an+1 =
{an, 2an − an−1 +
6 (1− δ) ρ (an)
β (3δ + (1− δ) ρ (an))
}. (140)
In general, the solution to the above difference equation, and the resulting equilibrium, de-pends on the specification for ρ (θ). However, for N = 1, the conditions simplify. Specifically,note that in this case, aN−1 = a0 = l, so that an equilibrium exists if there is a solution qLeakN
to q = g (q), where
g (q) ≡δ(q − β
4(q − l)2)
+ (1− δ)[
ρ (q){q+l2− β
12(q − l) 2
}+ (1− ρ (q))
(γ(h+l
2− β
12(h− l)2)+ (1− γ)
(q+l2− β
12(q − l)2)) ] .
(141)Note that
g (l) =
δl
+ (1− δ)[
ρ (l) l
+ (1− ρ (l))(γ(h+l
2− β
12(h− l)2)+ (1− γ)l
) ] (142)
but since h+l2− β
12(h− l)2 > l, we have g (l) ≥ l. Next, since
h− β4
(h− l)2 < h, and h+l2− β
12(h− l) 2 ≤ h, (143)
we have g (h) < h. This implies there always exists an equilibrium when N = 1.We numerically explore equilibria for N > 1 for a particular specification for ρ (θ). Specif-
ically, suppose
ρ (θ) = ρ
(h− θh− l
)∈ [0, ρ] , (144)
which implies ρND = ρh− q+l
2
h−l . Figure 5 plots the equilibrium disclosure threshold for thisspecification as a function of the maximum probability of a leak ρ. A comparison of this
34
plot to the one in Figure 1 suggests that the equilibrium threshold responds similarly to theprobability of leakage as before: the likelihood of disclosure is higher as the probability ofleakage increases, and more informative internal communication is more difficult to sustainin this case. While the numerical results are sensitive to the specification of ρ (θ), we expectthe qualitative implications of our analysis to be similar in this more general case.
Figure 5: Cutoff q as a function of ρ when ρ (θ) = ρ(h−θh−l
)The figure plots the disclosure threshold q, when (i) there are no leaks, and (ii) when thereare leaks for different values of N . Benchmark parameters are h = 2, l = 0, δ = 0.75, β = 1and π = 0.5.
0.2
0.4
0.6
0.00 0.25 0.50 0.75 1.00ρ
q
No LeakLeak, N = 1Leak, N = 2
35